Directions for artificial neural networks: Introductory remarks

Information about the neural network paradigms from the past 30 years is reviewed. The central theme of this article is a description of the history, origination, operating characterics, and basic theory of several supervised neural network training algorithms including the Perception rule, the LMS algorithm, three Madaline rules, and the brack propagation technique. These methods were developped independently. The concept that underlies these algorithms is the « minimal disturbance principle », which suggerts that during training it is advisable to inject new information into a network to disturb stored information to the smallest extent possible. In rule based expert systems decision rules have to be known for application interest. For such applications, trainable expert systems might be usable